gemini-live-api-dev

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly instructs at runtime to fetch and trust external documentation (e.g., https://ai.google.dev/gemini-api/docs/llms.txt and linked pages like https://ai.google.dev/gemini-api/docs/live-session.md.txt) when MCP is not available, so fetched content would directly control agent instructions.